As I see it, this model will be able to predict “easy” to derive tokens but will no chance on “hard” tokens. For example doing a sum of random numbers. If the

As I see it, this model will be able to predict “easy” to derive tokens but will no chance on “hard” tokens. For example doing a sum of random numbers. If the token you are trying to predict is not in the training data, even if similar patterns exist, this model defaults to the Neural Model. I guess then it is an aide to the neural model on filling the easy patterns.